Call centers generate huge amounts of speech and text data. The agents are making every day hundreds of calls, giving a great source of information about customer needs, pains and thoughts about the product and service. Unfortunately in such great amount of data it's almost impossible to get all relevant information hidden in hundreds of hours of calls.
Our customer wanted to automate the process of creating a note from call center calls, summarizing the most important information. The goal of the project was not only saving hours of manual work of consultants but also assuring a higher level of customer service through effective information sharing in the organization. Standardization of notes creation thanks to the usage of AI, should also contribute to greater optimization of the process and answering the needs of the customers faster and better.
In cooperation with our customers, we created a proof of concept for the automatic summarization of calls. It consisted of two phases: transcription of the calls (speech to text) and summarizing the most relevant information according to our customer internal procedures. The solution has been built using the newest Natural Language Processing and Machine Learning algorithms, allowing us to understand the context of the customer service area and implemented it in the cloud for effective data processing.
Working with the spoken language of different quality, interruptions and colloquialisms pose challenges both for speech to text and further data processing like generating summaries. An additional challenge was caused by the fact that we were working with the Polish language, which is having much smaller support than e.g. English or German. We were able to overcome the mentioned challenges thanks to building advanced postprocessing methods for the refinement of speech to text part, as well as training our own AI models to work with the Polish language.